@dhh I've lately been using GPT5.5 medium. I feel like choosing less thinking effort allows for much better steering and control of the output of the agent. Feels nicer than having x-large thinking and then trying to undo "the mess"
@louispereira Hands off as in: I don't look at the code it produces, I just case about the end result. I tent to accept permissions by default.
This is probably about 10-20% of the cases, for the rest I use cursor and check at the code diff.
@thekitze Careful with going through the openSCAD rabbit hole. 3D models based on code, so you can use parameters etc. Often what's behind the SkAPA build tool
@kepano Helping Claude Code understand โmetadataโ like rendered bases or custom skills to be able to read canvas files would bridge the gap. Claude code with obsidian is amazing already!
@kepano@obsdmd A config in bases to create a helper file with the rendered markdown of the bases. Helpful to use together with AI agents, that can't see the rendered bases result.
@dhh "Tolerance for computer bs".
I think same goes for LLMs: Need to develop tolerance for LLM bs if you dream of making something of yourself in technoligy.
@fortelabs Actually, doesn't work.
Develop of Obsidian copilot mentions that O series have a different API.
Will add them in next update: https://t.co/brOMtq5Kxm
@fortelabs In the settings of Copilot, you can go to Model tab and add custom model.
Model name: "o3-mini-2025-01-31"
And then just paste the OpenAI API key